AI-Generated Works and Copyright in India: What the DABUS Ruling Means for Authorship

• The application: Stephen Thaler sought registration for an artwork called “A Recent Entrance to Paradise”. The application identified DABUS—“Device for the Autonomous Bootstrapping of Unified Sentience”—as the author, while Thaler claimed ownership.

• Copyright Office’s objection: Indian copyright law does not merely require an “author” field to contain the name of whatever physically or computationally generated an output. The person entered as author must satisfy the legal definition contained in the Copyright Act.

• Opportunity to correct the application: During the proceedings, Thaler was specifically given an opportunity to identify himself as the author under Section 2(d)(vi). He declined because his legal position was that DABUS itself should receive authorship credit.

• Final outcome: The Registrar therefore reached an unusual but important combination of conclusions: the artwork could be original; DABUS could not be the legal author; Thaler was the legally recognised person capable of being treated as statutory author on these facts; but the application as actually maintained could not be registered.

• AI system: DABUS is an artificial-intelligence system developed by Stephen Thaler using interconnected artificial neural-network modules. According to the material placed before the Copyright Office, its internal process involved training, association formation, variation or “perturbation”, monitoring, reinforcement and output generation.

• Human inputs: Thaler stated that the visual material included photographs personally taken by him, while linguistic inputs included curated thesaurus entries and English sentences. He also supplied brief descriptions connecting visual and linguistic information.

• Autonomous final stage: According to Thaler, once the system had been configured and trained, the final artwork emerged without real-time human intervention and without a conventional text prompt. The Copyright Office accepted that DABUS performed the immediate computational generation, but held that this fact did not by itself decide legal authorship.

• UPSC significance: This demonstrates why “AI-generated” is not a single technological category. Human involvement may occur in system design, data selection, prompting, configuration, selection of outputs, editing or post-production, and the legal consequences can vary with the facts.

• Originality: This asks whether a work has the minimum independent expressive character required for copyright protection. It is primarily associated with Section 13.

• Authorship: This asks whom the Copyright Act legally treats as the creator of the work. For computer-generated literary, dramatic, musical or artistic works, Section 2(d)(vi) supplies a special rule.

• Ownership: This asks who holds the bundle of economic copyright rights. Under Section 17, the author is ordinarily the first owner, subject to statutory exceptions such as certain employment and commissioned-work situations. Rights may thereafter be assigned according to Sections 18 and 19.

• Why the distinction matters: A work may satisfy the originality threshold but still require identification of a legally recognised author and a valid chain of ownership. This is precisely why the DABUS application failed even after the artwork was found original.

• Special statutory rule: Section 2(d)(vi) defines the author, in relation to a computer-generated literary, dramatic, musical or artistic work, as “the person who causes the work to be created”. The provision was introduced through the Copyright Amendment Act, 1994, long before modern generative AI.

• Technology-neutral wording: Parliament did not say that the computer itself becomes author. Nor did it require the legally recognised author personally to execute every final pixel, note or word.

• Registrar’s interpretation: The focus is on the person responsible for bringing the particular work into existence. The immediate technological generator and the legal author can therefore be different.

• Important limitation: The Office expressly rejected a simplistic “machine owner” rule. Merely owning, developing or deploying a model cannot make someone author of all outputs. The causal connection must be established for the particular work.

• This distinction is extremely important for modern GenAI platforms, where the model developer, platform operator, user, prompt writer, employer and editor can all be different persons.

• Indian originality standard: The Copyright Act does not define “original”. The leading Supreme Court precedent relied upon by the Registrar is Eastern Book Company v. D.B. Modak, which requires at least a minimal degree of creativity rather than patent-like novelty or inventiveness.

• Copyright is not patent law: A copyrighted artwork does not have to be technologically new, non-obvious or revolutionary. Copyright protects original expression; patents apply a much higher novelty and inventive-step test to inventions.

• Application to DABUS: On the material before the Office, the final arrangement of forms, colours, tonal variations and spatial elements was not shown to reproduce an identified pre-existing work and was not simply the inevitable result of a mechanical command. It therefore crossed the limited originality threshold.

• Crucial qualification: The finding did not establish that every AI output is automatically original. Nor did it determine the lawfulness of every item that may have been used to train an AI system. It concerned one artwork and the evidence presented in one registration proceeding.

• No automatic immunity: Originality and infringement are separate inquiries. An AI-generated work may contain original elements while still raising infringement questions if protected expression from another work has been substantially reproduced.

• Expression, not mere idea: Copyright generally protects original expression rather than facts or abstract ideas. Therefore, similarity must be analysed in terms of legally protected expression.

• DABUS limitation: The Registrar found no clear evidence in this proceeding that “A Recent Entrance to Paradise” reproduced an identified earlier work. That should not be converted into a general finding that AI systems never reproduce copyrighted material.

• Legal personality: Indian law distinguishes between natural persons and recognised juristic persons. A company, for example, can possess legal personality because the law recognises it as capable of owning property, contracting, suing and being sued.

• AI is different: No existing Indian statute gives an AI software system such independent legal personality. DABUS therefore cannot acquire property simply because it can generate sophisticated output.

• Copyright carries legal consequences: Authorship is not merely a credit line. Copyright law is connected with ownership, licensing, assignment, enforcement, liability and special rights of authors. Treating software itself as author would affect this entire legal architecture.

• Administrative limit: The Registrar therefore refused to create a new type of juristic person through interpretation. Whether autonomous AI should ever receive such status is, according to the order, a legislative question.

• Direct nexus with the work: The conclusion did not rest merely on Thaler owning DABUS. He had created and configured the system, supplied the relevant visual material, curated linguistic material, linked those inputs through descriptions and initiated the process producing this particular artwork.

• Tool versus legal originator: DABUS carried out the final computational operation, but the Copyright Office distinguished the technological mechanism from the person to whom the creative process could legally be attributed.

• Work-specific test: This is perhaps the most important implication of the decision. The Registrar stated that ownership or development of a system alone cannot make someone author of every output. A direct and substantial relationship with the particular creation is necessary.

• Therefore, the DABUS ruling should not be read as saying that the company which creates a generative-AI model automatically owns everything produced through it.

• No automatic rule: The DABUS decision does not establish that every prompt writer is automatically the statutory author.

• Degree of involvement matters: A person who iteratively designs prompts, selects reference material, adjusts parameters, curates alternatives, modifies outputs and integrates them into a larger human work presents a much stronger case for a work-specific causal connection than someone who types a one-line instruction and accepts the first output.

• Different factual situations: A future dispute may involve the model developer, employer, commissioning party, user or several contributors. Section 2(d)(vi) will have to be applied to the particular creation rather than mechanically awarding rights to the nearest human.

• Contracts are not the whole answer: A platform’s terms of service may allocate contractual rights between platform and user, but a private contract cannot make an AI system a statutory author or manufacture copyright where the requirements of copyright law are not satisfied.

• AI-assisted creation: A human may use AI like a sophisticated editing, translation, colour-correction, research or design tool while retaining control over the expressive choices. Conventional human authorship is easier to identify in such cases.

• Substantially AI-generated creation: The system may determine much of the final expressive output. The legal problem then becomes identifying who, if anyone, satisfies Section 2(d)(vi)’s causal attribution rule.

• Continuum rather than binary: Real-world works fall across a spectrum. A film, advertisement, game or book may contain human-written, AI-generated and heavily edited components. The DABUS order itself cautioned that different AI systems and works can involve materially different degrees of human participation.

• Policy implication: India may eventually require clearer disclosure rules so that registrars and courts can distinguish human expression, AI-generated elements and subsequent human modifications.

• Defective authorship chain: The application continued to identify DABUS as the author and Thaler as owner.

• Ownership problem: Under the statutory structure, the author is ordinarily the first owner under Section 17. If another person claims ownership, there must be an applicable statutory exception or a legally valid transfer. Because DABUS was incapable of holding copyright, it was equally incapable of assigning that copyright to Thaler under Sections 18 and 19.

• Refusal to amend: Thaler was offered an opportunity to identify himself as author but chose not to do so.

• Effect of rejection: The order specifically says rejection does not prevent him from seeking an available remedy based on legally correct authorship and ownership particulars. Thus, the case is not equivalent to a declaration that the artwork can never receive copyright protection.

• Automatic protection: No. The Copyright Office’s official handbook states that acquisition of copyright is automatic and registration is not a condition for the existence of copyright.

• Why register then?: Registration creates useful evidentiary value. Under Section 48, entries in the Register of Copyrights are prima facie evidence of the particulars recorded there.

• Registration procedure: Applications are governed by Section 45 and Rule 70. Form XIV is used for registration, while the Register itself includes separate fields concerning the author, applicant and owners of rights.

• Importance for DABUS: The case was formally a registration proceeding. Its wider importance comes from the Registrar’s detailed interpretation of how the existing Act applies to computer-generated works.

• No—this is a different question: The DABUS case concerns the copyright status and authorship of an output. AI training concerns the legality of using existing copyrighted works as inputs.

• Indian litigation: In ANI Media Pvt. Ltd. v. OpenAI OpCo LLC, the Delhi High Court, while deciding an application for interim injunction, held prima facie that OpenAI’s storage of ANI’s works for training its LLMs fell within Section 52(1)(a)’s fair-dealing framework in the circumstances before the Court. It also found, at that interim stage, that the demonstrated ChatGPT outputs were not substantial reproductions of ANI’s works. The observations are prima facie and the underlying suit is not equivalent to a final blanket rule for every AI-training situation.

• Two separate policy questions: India must therefore deal with both:

• whether developers may use copyrighted material to train AI; and

who receives rights over the material AI subsequently generates.

• Resolving one does not automatically resolve the other.

• Government’s 2024 position: In a Rajya Sabha reply, the Ministry of Commerce and Industry stated that the existing IPR framework was considered capable of dealing with AI-generated works and that there was then no proposal for a separate category of AI-specific rights. It also emphasised that IPR owners are legal persons.

• How DABUS fits this position: The Copyright Office did not interpret the statement as recognising AI itself as author. Instead, it read existing law as capable of accommodating computer-generated works by attributing authorship to a legally recognised person under Section 2(d)(vi).

• Evolving policy: The position is not frozen. India’s rapidly expanding use of generative AI has subsequently led to deeper examination of training data, output ownership, liability and creator compensation.

• Expert committee: DPIIT constituted an eight-member committee to examine whether existing copyright law adequately addresses generative AI and whether amendments are required.

• Part I: The first working paper concentrates principally on the input or training side. It examines approaches such as text-and-data-mining exceptions, voluntary licensing and collective licensing and proposes a hybrid framework involving access to lawfully obtained content and remuneration after commercialisation. This remains a policy proposal rather than enacted copyright law.

• Output questions: Questions such as copyrightability of AI outputs, authorship, moral rights and liability have been identified for separate examination. The DABUS order therefore arrives at an important time in India’s broader AI-copyright policy debate.

• Need for coordination: Any future output-side framework will have to be harmonised with the Copyright Act, AI governance policy and the emerging judicial interpretation of AI training and fair dealing.

• United States position: The U.S. Copyright Office’s 2025 report concluded that generative-AI outputs can receive copyright protection only where a human author has determined sufficient expressive elements. AI assistance does not destroy human copyright, but prompts by themselves generally do not establish copyright in machine-determined expressive output.

• Thaler litigation in the US: Stephen Thaler separately attempted to obtain U.S. copyright recognition for “A Recent Entrance to Paradise” with the Creativity Machine identified as creator. U.S. courts maintained the human-authorship requirement. The U.S. Supreme Court denied his petition for review in March 2026, leaving the lower-court result undisturbed.

• Difference from India: India has an express statutory provision for computer-generated works in Section 2(d)(vi). Consequently, the Indian approach can recognise an original computer-generated output while legally attributing its authorship to the person who caused it to be created. That makes India’s statutory framework structurally different from the U.S. human-authorship approach.

• Statutory attribution: Section 9(3) of the UK Copyright, Designs and Patents Act, 1988 provides that for a computer-generated literary, dramatic, musical or artistic work, the author is taken to be the person “by whom the arrangements necessary for the creation of the work are undertaken”.

• Similarity with India: The UK approach resembles India’s Section 2(d)(vi) because both statutes look beyond the machine executing the final operation and attribute authorship to a person connected with bringing the work into existence.

• Global diversity: WIPO notes that only a limited group of jurisdictions, including India and the UK, expressly provide for computer-generated works in this fashion. Other jurisdictions rely more strongly on direct human creative contribution.

• The absence of a uniform international model creates complications for cross-border licensing and enforcement of AI-generated content.

• Legislative competence: Patents, inventions, designs, copyright and trademarks fall under Entry 49 of the Union List in the Seventh Schedule to the Constitution. Parliament therefore has legislative competence over copyright law under the constitutional distribution of powers.

• Institutional framework: The Copyright Office operates under DPIIT in the Ministry of Commerce and Industry and administers copyright registration under the Copyright Act and Copyright Rules.

• Separation of institutional roles: The Registrar can interpret and administer existing law, while courts can adjudicate disputes and review legal questions. Creation of an entirely new legal status for autonomous AI, however, would require Parliament rather than an administrative registration decision. This institutional distinction is central to the DABUS order.

• Greater possibility of protection: Artists, filmmakers, advertisers, game developers, publishers and musicians should not assume that using AI automatically destroys copyright. Human-AI works can remain protectable depending on originality and statutory authorship.

• Need to document human contribution: Creators using generative tools will increasingly benefit from retaining evidence of prompts, source material, iterations, selections, edits and final human modifications.

• Chain-of-title concerns: Film studios, publishers and businesses purchasing AI-assisted content will need certainty regarding who created the output and who possesses authority to license it.

• Contract risk: Commercial agreements should distinguish ownership of human-created input, prompts, AI output, edited output and underlying third-party material.

• Competition implications: Over-generous rights could allow firms to monopolise large quantities of automatically generated expression, while denying protection altogether could reduce incentives for legitimate AI-assisted creative investment. Copyright policy therefore requires a balance between innovation, competition and creator incentives.

• Mass generation: Can a person realistically be treated as author of millions of automatically generated outputs merely because that person deployed a system? The DABUS order’s work-specific approach points against such an automatic conclusion.

• Moral rights: AI authorship would create difficult questions under provisions such as Section 57, because attribution and integrity rights are structured around legally recognised authors.

• Ownership versus contractual licences: Platform terms can regulate relations between parties but do not conclusively answer statutory authorship.

• Infringing outputs: Liability remains important where AI-generated output substantially reproduces protected expression.

• Training data: The relationship between Section 52, licensing, text-and-data mining and remuneration for creators remains a major policy issue.

• Cross-border recognition: An AI-assisted work treated as copyrightable in India may face a different result in a jurisdiction applying a stricter human-authorship test.

• Provenance: Copyright authorities may increasingly require reliable information about how AI-assisted works were actually produced.

• GS3—Science and Technology: It illustrates how technological innovation can outpace legal categories and require technology-neutral interpretation.

• GS3—Intellectual Property Rights: The case directly connects copyright, originality, authorship, ownership and computer-generated works.

• Governance dimension: It raises the question of how regulators should respond to emerging technology without assuming powers that properly belong to Parliament.

• Economic dimension: Clear AI-copyright rules are important for India’s creative economy, startups, software companies, publishing, entertainment, media and AI innovation.

• Ethical dimension: The debate concerns appropriate credit for human creativity, transparency about machine-generated content and accountability for autonomous technological processes.

• Mains perspective: The larger issue is not simply whether “AI deserves copyright”, but how law can allocate incentives and responsibility when human creativity and machine generation increasingly interact.

• Parliament and DPIIT should provide greater clarity on the factors relevant to Section 2(d)(vi), including prompting, curation, model configuration, output selection, editing and other work-specific contributions, while retaining enough flexibility for technological change.

• Copyright registration procedures could require proportionate AI provenance disclosure—whether AI was used, the broad nature of human contribution and whether substantial AI-generated material forms part of the work—without treating every ordinary digital tool as generative AI.

• India should maintain a clear distinction between AI-assisted works, computer-generated works and situations where meaningful legal authorship cannot be established. A single rule for all three would create uncertainty.

• The output-side question of authorship should be coordinated with the Government’s ongoing framework for training data, licensing, fair dealing and creator remuneration. The rights granted over AI output should not be designed independently of the rules governing the material used to train AI.

• Legal certainty is needed regarding the respective roles of users, developers, employers, commissioning parties and platform operators. Default statutory rules can reduce disputes while allowing lawful contractual arrangements to operate.

• India should avoid granting AI independent legal personality merely to solve copyright attribution. Rights can presently be assigned to accountable natural or recognised juristic persons; creating AI personhood would have consequences far beyond copyright, including contracts, property, liability and taxation.

• Registration systems and courts should encourage preservation of prompt histories, editing records and other reasonable evidence of human contribution where authorship is disputed. Such records can improve transparency without imposing excessive compliance costs on ordinary creators.

• International engagement through WIPO and dialogue with major jurisdictions will be important because AI-generated works, platforms and markets are inherently cross-border while national copyright rules remain different.

• Policymakers should balance creator incentives with competition and the public domain. Over-protection of automatically generated material could create enormous artificial monopolies, while excessive uncertainty could discourage legitimate investment in human-AI creativity.

• The law should remain technology-neutral wherever possible. Rather than focusing on whether an output “looks as if” a human or AI made it, copyright analysis should continue to examine originality, statutory authorship, lawful ownership, infringement and accountability separately.